Enabling Memory Hot Add forces the hypervisor to present a single interleaved virtual NUMA node (UMA) to the guest OS. This occurs because ESXi must maintain flexibility for future memory expansion, which prevents the static optimization required for vNUMA.
In vSphere 8.0 and later (Hardware Version 20+), CPU Hot Add has been improved to maintain vNUMA topology (using numa.allowHotadd = TRUE), but Memory Hot Add still completely deactivates vNUMA.
Following two scenario may also cause this issue as well:
Note: Enabling this feature disables vNUMA, which can impact the performance of large virtual machines (typically those with 8 or more vCPUs).